RIKEN’s second notice on unauthorized access to the Fugaku and RIKYU supercomputers turns an alarming but abstract cyber incident into something much more operational: 4,222 Fugaku users now have potentially compromised registration data, and 430 authentication credentials stored in user home directories were collected and exfiltrated. For research institutions, cloud-HPC providers, and businesses that rely on shared compute environments, the immediate question is not only how the attacker got in. It is how quickly operators can find every key, token, and trust path that might let one compromised cluster become a bridge into many other systems.
In the second notice, RIKEN said the intruder accessed targeted information relating to 4,222 Fugaku users, including addresses, affiliations, demographic information, phone and other contact details, and records showing whether identity verification had been completed. It said it has not confirmed leakage of the identity documents themselves. More consequential for the wider ecosystem, RIKEN found evidence that 430 credentials used to connect from user home directories to external systems were collected and exfiltrated: 233 private keys without passphrases and 197 private keys with passphrases. Some of those credentials were used in attempts to connect to external systems, RIKEN said, though it has not identified other cases of misuse.
Fugaku will not resume until RIKEN completes steps including separating public services from the administrative network and rebuilding servers. The organization said it preserved evidence, notified police in Saitama and Hyogo on November 16 and 19, and is contacting affected users individually to revoke and reissue credentials.
Why the key count matters more than the machine count
Fugaku and RIKYU are not isolated corporate servers. They are shared national research resources used by universities, public labs, and private-sector researchers. That means the most important assets on the compromised systems may not be the supercomputers themselves, but the identities they mediate. A private key sitting in a user home directory can authorize access to another supercomputer, a cloud account, a code repository, a lab instrument, or an institutional system somewhere else.
That is why this incident looks less like a contained break-in and more like a supply-chain-style identity event. If a long-lived credential can be reused from a shared environment, the blast radius is defined by every place that credential can reach. The 233 keys without passphrases are the clearest example: if still valid, they reduce the attacker’s next move to a routing problem. Even the 197 passphrase-bearing keys should not be dismissed. RIKEN has not said how those passphrases were managed in practice, and some of the collected credentials were already used in attempted external connections.
Independent reporting from BleepingComputer described the public operational-status dashboard as the apparent springboard for the intrusion between November 10 and 13, followed by movement across multiple internal servers and malware aimed at user data. RIKEN is still clarifying the circumstances, so the route in matters less right now than the route out that the stolen credentials may have created.
What operators should do before a compromise spreads
The practical lesson is straightforward: treat every credential stored on a shared compute system as part of your attack surface, not as a user convenience.
First, inventory every key and token on shared systems and map where each one can authenticate. Many operators know which accounts exist on a cluster but not which outside services those accounts can reach. That gap becomes costly as soon as a home directory is exposed.
Second, assume exposed credentials are compromised even if you have not confirmed successful misuse. RIKEN’s notice is a case study in why waiting for proof can backfire. The credentials were collected and exfiltrated, and some were used in attempts to connect externally. Revocation should happen at every destination system, not just on the breached cluster.
Third, reissue access with tighter limits. Short-lived, scoped credentials reduce the value of anything left behind in a shared environment. Where possible, remove private keys from home directories entirely and use brokered access, hardware-backed credentials, or just-in-time authentication instead of long-lived SSH material.
Fourth, separate what must face the internet from what administers the cluster. RIKEN’s decision to keep Fugaku offline until public services are separated from the administrative network is one of the most important details in its update. Shared HPC environments often let public dashboards, schedulers, storage, user space, and management networks sit too close together. Recovery should shrink those trust relationships, not merely recreate them on clean hardware.
Fifth, watch for the quiet signs of spread: outbound authentication attempts, new trust relationships, unusual east-west traffic, and changes to login scripts, binaries, or job workflows. In research computing, integrity matters as much as confidentiality. An attacker who can alter software stacks or datasets can damage science and business results without ever stealing a file.
The real recovery test
The biggest business lesson in the RIKEN notice is that recovery quality matters more than dramatic headlines. A strong response is not measured only by whether malware was removed. It is measured by whether an operator can name the affected identities, prove which external systems were reachable, revoke access without guessing, rebuild from clean images, and tell users exactly what they must rotate.
RIKEN has become more specific on the first two counts than many institutions do in public. It has put concrete numbers on affected users and exfiltrated credentials, and it has acknowledged that the system will need segmentation changes and rebuilt servers before trust can be restored. That is the right direction.
But important unknowns remain. RIKEN has not publicly identified the attacker, the full list of compromised servers, the exact initial vulnerability, the specific outside systems contacted, whether research files were exfiltrated, or when Fugaku and RIKYU will return to normal service. It has not confirmed that research results were stolen or that outside systems were successfully accessed with the collected credentials. And 4,222 affected Fugaku users should not be read as 4,222 compromised external systems; users have different permissions, different habits, and different credential hygiene.
Still, the operational answer for peers is already clear. If your shared environment can store long-lived credentials that open doors elsewhere, you have to be able to enumerate them, rotate them fast, narrow their scope, and restore from a design that limits what one compromised service can touch. RIKEN’s second notice makes that requirement concrete.




By
By
By
By
By








